COLO-HEP-288

hepth@xxx/9207095
July 1992; Revised October 1992

Quantum Black Holes in Two Dimensions

arXiv:hep-th/9207095 v2 14 Oct 1992

⋆
S.P. de Alwis

Dept. of Physics, Box 390,

University of Colorado,
Boulder, CO 80309

ABSTRACT

We show that a whole class of quantum actions for dilaton-gravity, which reduce
to the CGHS theory in the classical limit, can be written as a Liouville-like theory.
In a sub-class of this, the field space singularity observed by several authors is
absent, regardless of the number of matter fields, and in addition it is such that
the dilaton-gravity functional integration range (the real line) transforms into itself
for the Liouville theory fields. We also discuss some problems associated with
the usual calculation of Hawking radiation, which stem from the neglect of back
reaction. We give an alternative argument incorporating back reaction but find
that the rate is still asymptotically constant. The latter is due to the fact that
the quantum theory does not seem to have a lower bound in energy and Hawking
radiation takes positive Bondi (or ADM) mass solutions to arbitrarily large negative
mass.

⋆ dealwis@gopika.colorado.edu
 1. Introduction

The theory of dilaton gravity coupled to scalar fields proposed by Callan et al,
[1] (CGHS) has generated a flurry of activity on black hole physics. What one has
is a simple toy model, within which the puzzling questions associated with Hawking
radiation [2] can be addressed in a systematic way. In the original work of CGHS as
well as in several subsequent papers, it was assumed that quantum effects to leading
order could be included by just adding a piece to the action which reproduced the
conformal anomaly. However it was later realized that the consistent quantization
of the theory in conformal gauge, required that the cosmological constant term
and/or the kinetic terms should get renormalized in a dilaton dependent manner,
so that the theory becomes a conformal field theory (cft)[3, 4]. This requirement
that the theory be an exact cft (though not necessarily a soluble one) is not a
matter of choice. It is a necessary consequence of general covariance. In other
words dilaton gravity coupled to matter fields must be a cft in exactly the same
way that string theory (i.e ordinary 2d gravity coupled to matter fields) is a cft.

In this paper we will first review this argument and then consider the general-
ization of previous solutions to the conformal invariance conditions. We show that
there is a subset of models which are free of the quantum black hole singularity
pointed out in [5, 6], and which are such that the original range of integration for
the conformal factor and the dilaton, is transformed into itself for the Liouville the-
ory fields. We will also argue that the calculations of Hawking radiation that have
been given in the literature, are inconsistent with the constraints and equations of
motion of the theory since they neglect back reaction. There is no sensible approx-
imation scheme in which the latter can be ignored. We then show that when the
exact solution of the system of equations coming from quantum corrected action
is considered, the results differ from previous calculations. However it turns out
that one cannot see the radiation turning off in this theory, the (Bondi) mass of
the solutions of the theory can be arbitrarily negative, and the Hawking process
causes a positive mass solution to decay indefinitely to infinitely negative mass.

2
Altough Liouvile theory has a positive definite spectrum the same is not true of
the Liouville-like theory that is obtained from the CGHS theory. It is possible that
the origin of the problem lies in the the CGHS theory itself, but a more rigorous
quantum treatment of the Bondi mass may resolve this question.

In the next section we review the quantization of the CGHS theory. In the
third section we discuss a class of solutions to the integrability conditions for the
constraints and present arguments for taking the resulting exact conformal field
theory as a quantum theory of dilaton gravity. In the fourth section we demonstrate
explicitly how the classical singularities are tamed by quantum effects. In the fifth
section we review the CGHS calculation of Hawking radiation in this model. In
the sixth section we give an alternative calculation which is consistent with the
constraints (this is basically a detailed version of a calculation contained in the
second paper of [3]) and in the final section we make some concluding remarks.

2. Quantization

The CGHS theory is defined by the classical action

N
1 √ 1X
Z
2 −2φ 2 2
S= d σ −g[e (R + 4(∇φ) + 4λ ) − (∇f i )2 ]. (2.1)
4π 2
i=1

In the above G is the 2d metric, R is its curvature scalar, φ is the dilaton and
the f i are N scalar matter fields. This action may be obtained as a low energy
effective action from string theory, in which case the f fields will arise from the
Ramond-Ramond sector. Note that the (zero mass) tachyon of 2d string theory is
excluded from this action. If this field had been coupled then the theory would
⋆
not be solvable even at the classical level.

The quantum field theory of this classical action may be defined as

⋆ For a discussion of how in this case, 2d black hole solutions are affected far away from the
black hole by the presence of the tachyon, see [7].

3
 [dg]g [dφ]g [df ]g iS[g,φ,f ]
Z
Z= e . (2.2)
[V ol. Dif f.]

The metrics which define these measures are usually given by,

√
Z
||δg||2gd2 σ −gg αγ g βδ (δgαβ δgγδ + δgαγ δgβδ )
=
(2.3)
2 √ √
Z Z
||δφ||g = d σ −gδφ , ||δf ||g = d2 σ −gδij δf i δf j .
2 2 2

However we can be more general in these definitions as long as 2d diffeomorphism

invariance is preserved. Now let us gauge fix to the conformal gauge g = e2ρ ĝ and
rewrite the measures with respect to the fiducial metric ĝ. Following the work of
David and of Distler and Kawai [8], we may expect the action to get renormalized,
except that unlike in their case the renormalization will be dilaton dependent (since
the coupling is e2φ ). Thus in general we may expect the gauge fixed path integral
†
to be written as [9,7,3] ,

Z
Z= [dX µ ]ĝ [df ]ĝ ([db][dc])ĝ eiI(X,ĝ)+iS(f,ĝ)+iS(b,c,ĝ) , (2.4)

where

1 1
Z p
I[X, ĝ] = − −ĝ[ ĝ ab Gµν ∂a X µ ∂b X ν + R̂Φ(X) + T (X)]. (2.5)
4π 2

S(b, c, ĝ) is the Fadeev-Popov ghost action, and we have written (φ, ρ) = X µ . Note
that all the measures in (2.4) are defined with respect to the 2d metric ĝ and that
in particular the measure [dX µ ] is derived from the natural metric on the space
√
||δXµ ||2 = d2 σ −ĝGµν δX µ δX ν . In the limit of weak coupling (e2φ << 1) we
R

have,

† For alternative approaches to the quantization see [10].

4
 1
Z
d2 σ ˆ 2 − 4∇φ.
−ĝ[e−2φ (4(∇φ) ˆ ∇ρ)
ˆ − κ∇ρ.
ˆ ∇ρˆ
p
I→
4π (2.6)
−2φ 2 2(ρ−φ)
+ R̂(e − κρ) − 4λ e ]

This is obtained from (2.1) by putting g = e2ρ ĝ, and including a very specific
higher order term; namely the usual conformal anomaly term. κ in the above is
26−(N +2) 24−N
equal to 6 = 6 , if one includes the contribution of the transformation
‡
of the measure for φ and ρ.

I is a generalized sigma model action and we have kept only renormalizable

terms. The sigma model action introduces three (dilaton dependent) coupling
functions G, Φ, and T , respectively the field space metric, dilaton, and tachyon.
The only a priori restriction arises from the fact that the functional integral for Z in
(2.4), must be independent of the fiducial metric ĝ, as is obvious from the expression
(2.2) for it. This implies that the following constraints should be satisfied:

< T±± + t±± >= 0, (2.7)

and

< T+− + t+− >= 0, (2.8)

where Tµν is the stress tensor for the dilaton-gravity and matter sectors, and tµν
is the stress tensor for the ghost sector. ((2.8) is equivalent to the equation of
motion for ρ, and so is not an additional constraint). Furthermore one has to
satisfy the integrability conditions for these constraints, namely that they generate
§
a Virasoro algebra with zero central charge. As is well known (see for instance [11]
and references therein) this is equivalent to the requirement that the β-functions
[12] corresponding to the coupling functions, G, Φ, and T , vanish.

‡ We will justify this in more detail later on.

§ In effect this means that the field space must be exactly like the target space of string theory,
though here we do not give this space a space-time interpretation. The only space-time in
the theory is the original one parametrized by the coordinates σ.

5
 βµν = Rµν + 2∇G
µ ∂ν Φ − ∂µ T ∂ν T + . . . ,
(N + 2) − 26
βΦ = −R + 4Gµν ∂µ Φ∂ν Φ − 4∇2G Φ + + Gµν ∂µ T ∂ν T − 2T 2 + . . . ,
3
βT = −2∇2G T + 4Gµν ∂µ Φ∂ν T − 4T + . . . ,
(2.9)

where R is the curvature of the metric G. These equations have to be solved

under the boundary conditions that in the weak coupling limit (e2φ << 1) we get,
comparing (2.5) with (2.6),

Gφφ = −8e−2φ , Gφρ = 4e−2φ , Gρρ = 2κ, (2.10)

Φ = −e−2φ + κρ, T = −4λ2 e2(ρ−φ) . (2.11)

3. From CGHS to Liouville

Let us first discuss the renormalization of the field space metric and dilaton
(G and Φ) and postpone the discussion of the tachyon T . The (renormalized) field
space metric may be parametrized as,

ds2 = −8e−2φ (1 + h(φ))dφ2 + 8e−2φ (1 + h(φ))dρdφ + 2κ(1 + h)dρ2 , (3.1)

where h, h, and h are O(e2φ ). If we are going to consider only O(e2φ ) effects then
we should certainly set h to zero. But even if we consider the renormalization
functions h and h to all orders, it is consistent to limit ourselves to the class of
quantum versions of the CGHS theory which have h = 0, provided that we satisfy
the beta function equations. This corresponds to confining ourselves to theories in

6
which the field space curvature R = 0. In this case we can transform this metric
¶
to Minkowski form. First put

2
Z
−1 −2φ
y =ρ−κ e + dφe−2φ h(φ). (3.2)
κ

Then the metric becomes

8
ds2 = − P 2 (φ)dφ2 + 2κdy 2,
κ

where

1
P (φ) = e−2φ [(1 + h)2 + κe2φ (1 + h)] 2 . (3.3)

Putting

Z
x= dφP (φ), (3.4)

we have

8
ds2 = − dx2 + 2κdy 2 (3.5)
κ

With this form of the metric, ignoring O(T 2 ) terms, we find from the first
(graviton) β-function equation in (2.9), that ∂µ ∂ν Φ = 0. In other words Φ is
linear in x, y. Demanding that we recover the CGHS Φ given in (2.11) in the weak
coupling limit we find the unique solution,

¶ We will only consider theories with κ 6= 0 i.e. N 6= 24.

7
 Φ = κy. (3.6)

Substituting in the second (dilaton) equation in (2.9), we then get

24 − N
κ= . (3.7)
6

To determine T we consider the third equation of (2.9), to linear order and get,

κ 2 1
∂x T − ∂y2 T + 2∂y T − 4T = 0. (3.8)
4 k

This has solutions of the form T = eβx+αy , where κ 2

4β − k1 α2 + 2α − 4 = 0.
Now we need to impose the boundary condition that we recover the CGHS tachyon
given in (2.11) in the weak coupling limit. To do so we expand the expression for
x ((3.4), (3.3)) to get

1
Z
κ
x ≃ − e−2φ + dφe−2φ h + φ + O(e2φ ). (3.9)
2 2

Then we find that − κ4 x + 2y = 2ρ − 2φ + O(e2φ ) so that the unique solution (con-

fining ourselves to multiplicative renormalizations) obeying the required boundary
condition is
4
T = −4λ2 e− κ x+2y . (3.10)

⋆
For κ > 0 there is another (additive) term satisfying the boundary condition.
Namely
√4 2
Tnp = µe κx ≃ µ exp(− √ e−2φ ).
κ

This is in fact a non-perturbative ambiguity. We will set µ = 0 in the rest of

⋆ I wish to thank Andy Strominger for pointing this out to me [13].

8
the paper. In any case it is absent for κ < 0, since in that case we will have an
oscillatory solution which will not vannish in the classical limit.

It is convenient now to introduce rescaled fields,

s
2 p
X=2 x, Y = 2|κ|y, (3.11)
|κ|

in terms of which the metric and the tachyon become,

ds2 = ∓ dX 2 ± dY 2
q
2
2 ∓ |κ|
(X∓Y )
T = − 4λ e

In the above and in the equations below, upper/lower signs correspond to having
κ > 0/κ < 0 respectively. In terms of the new field variables the functional integral
becomes,

Z
Z= [dX][dY ][df ][db][dc]eiS[X,Y,f ]+iSghost , (3.12)

where,

q
1
Z 2
2 ∓ (X∓Y )
X
S= d2 σ[∓∂+ X∂− X ± ∂+ Y ∂− Y + ∂+ f i ∂− f i + 2λ e |κ|
]. (3.13)
4π
i

Several comments need to be made about this functional integral. First and
most obviously there is the question of the range of the integration. As we see from
(3.4) and (3.3), in general the range of integration in X will not extend over the
whole real line. What we then have is an approximate solution to the β-function
equations (2.9) valid only to leading order in the sigma model (α′ ) expansion and to
leading order in the weak field expansion in T . On the other hand if we define the

9
quantum theory by (3.12) with the range of integration for X being the whole real
†
line, we have a solution to the exact β-function equations. Thus this definition of
quantum dilaton-gravity theory, even if somewhat unorthodox, is a very compelling
one. It is on the same footing as for instance the definition of 2 + 1 dimensional
quantum gravity given by Witten[14] in which the functional integral is taken over
all values of the vielbein field. Also let us point out that if we restrict the range
of integration to be consistent with the original definition of the quantum theory
then, since we only have a solution to the leading order β-function equations, it
seems as if we will need an infinite number of terms to satisfy the exact conformal
invariance conditions. It is plausible to suppose that this theory is equivalent to
the one above with the unrestricted range of integration. This argument is also
reinforced by the fact that, as in the usual Liouville theory, the integration range
is effectively cut off (albeit softly) by the Liouville potential term. Finally (and
perhaps this is the most compelling reason for the quantum Liouville-like conformal
field theory) there exist choices of h and h̄, for which when the integration ranges
for φ and ρ are as usual taken over the whole real line, the same is true for the
ranges for X and Y (see case d at the end of this section).

The second comment is with regard to the approximation in which the dilaton
and graviton loops can be ignored. By rescaling and translating the fields X, Y it is
easily seen that h̄ = κ so that the semiclassical approximation is valid only for large
κ. Thus one might be inclined to believe that any (even qualitative) conclusions
derived for the N < 24 theory[3] are drastically effected by dilaton graviton loop
corrections. On the other hand for N = 1 we get κ = 3.8 which is of the same
order as the relevant parameter in QCD where the approximation works quite well.

Finally we comment on the different possibilities for the functions h and h.

Three special cases have so far been discussed in the literature.

a) h = h = 0. i.e. the field-space metric of the classical CGHS Lagrangian

† The theory is very much like Liouville theory which is an exact cft [15]. In fact it is less
singular than Liouville. So one expects it to be an exact cft as well.

10
is not renormalized. However in this case the cosmological constant term T is
renormalized.

b)h = −e2φ , h = −2e2φ . This is the case proposed by Strominger [16]. In this
case both the metric G and the tachyon T are renormalized.

c) h = 0, h = − κ4 e4φ . This is the case considered in [17] where P 2 is a perfect

square (see (3.3) from which we find P = e−2φ (1 + κ4 e2φ )). In this case the metric
is (obviously) renormalized but the tachyon is not (as is easily seen from (3.10)
and the expressions for x and y with the above value of P ).

d) In all of the above cases the transformation (3.4) has a singularity when
κ < 0. For instance in case a) it is at e2φ = −κ−1 . It is however quite easy to find
a class of models which have no such singularity. Put h = ae2φ and h = be2φ . Then
putting e2φ = z the condition for the absence of a singularity is that the quadratic
equation z 2 P 2 = (a2 + κb)z 2 + (2a + κ)z + 1 = 0 has no real roots. i.e. we must
choose κ2 + 4(a − b)κ < 0. Obviously there are many solutions to these conditions
but one particular class is of particular importance since members of it naturally
allow the range of integration in the X, Y variables to go over the whole real line.
The simplest member of this class has h = 0 and h̄ = − κ2 e2φ . In this case we have
from (3.3),(3.4),(3.2),

κ2 4φ 1
Z
x= dφe−2φ (1 +
e )2
4
(3.14)
1 κ2
 
−4φ 21 |κ| −1 |κ| 2φ
=− ( +e ) + sinh e ,
2 4 4 2

and

y = ρ − κ−1 e−2φ − φ. (3.15)

Clearly as φ, ρ, range from −∞ to +∞ so do x and y.

11
 4. Exact solutions
⋆
The equations of motion coming from (3.13) are as follows.

∂+ ∂− f = 0, (4.1)

s q
2 2 2
(X+Y )
∂+ ∂− X =λ e |κ| ,
|κ|
s q (4.2)
2 2 2
(X+Y )
∂+ ∂− Y = − λ e |κ| .
|κ|

We have taken the case with the lower signs in (3.13) so that the discussion is
for N > 24. There is no qualitative difference in the other case so it is unnecessary
†
to write it out explicitly. These equations are easily solved. From (4.2) we have
q
∂+ ∂− (X +Y ) = 0, so that X +Y = |κ| + −
2 (g+ (σ )+g− (σ )), where g± are arbitrary
chiral functions. Substituting into the X equation of motion and integrating we
have

+ −
s s Zσ Zσ
2 2 + −
X =− (u+ (σ + ) + u− (σ − )) + λ2 dσ + eg+ (σ )
dσ − eg− (σ )
|κ| |κ| (4.3)
r
|κ|
= −Y + (g+ + g− ),
2

where u± are arbitrary chiral functions to be determined by the boundary condi-

tions.

By a coordinate choice we can set g± = 0. In these coordinates (the analog of

Kruskal-Szekeres coordinates for the black hole) we get

⋆ In this section and in section 6, wherever it is appropriate, all equations are to be understood
as being valid inside the functional integral, i.e. as expectation values of quantum operators.
Since following the arguments of reference [15] the theory can be mapped into a free theory
it is plausible that the only quantum effects come from normal ordering.
† It is also contained in [4] and the second paper of [3].

12
 s
2
X = −Y = − (u − λ2 σ + σ − ). (4.4)
|κ|

where u = u+ + u− .

These solutions are of course the same as those of CGHS, except that they are
for X and Y and all the effects of the quantum anomalies are now incorporated in
the expressions for them in terms of ρ and φ. To be explicit consider the case d)
discussed at the end of the last section (h = 0, h̄ = − κ2 e2φ );

s
2 κ2 4φ 1
Z
X =2 dφe−2φ [1 + e ]2
|κ| 4
4
Z
1
= 2|κ| dφ[1 + 2 e−4φ ] 2 ,
p
κ

and
s
p 2 −2φ p
Y = 2|κ|ρ + e − 2|κ|φ.
|κ|

In the weak coupling limit (e2φ << 1) we have from (4.4) the classical solution

e−2φ = e−2ρ = u − λ2 σ + σ − , (4.5)

which exhibits the classical (black hole type) singularity on the curve where
the right hand side vanishes. But the singularity is in the strong coupling region
where we have to use the strong coupling expansion (from the second line of the
above equation for X)
p e−4φ
X≃ 2|κ|[φ − + . . .].
κ2

Then we have from (4.4),

13
 φ ≃ κ−1 (u − λ2 σ + σ − ),

and
1 −2κ−1 (u−λ2 σ + σ − )
ρ≃ e .
κ

The metric (e2ρ ) is clearly non-singular at the classical singularity.

Differentiating the solution for X with respect to σ± we get

(∂± u± − λ2 σ ∓ )
2e−2φ ∂± φ = − , (4.6)
P̄ (φ)

where P̄ = e2φ P , P being defined by (3.3). This equation gives the trajectory
of the apparent horizon (∂+ φ = 0) introduced in [6] (once the unknown function u
is determined) as

1
σ− = ∂+ u+ (σ + ). (4.7)
λ2

By differentiating the solution for Y and using (4.6) and the expression for Y
in terms of ρ, φ, we have

κ∂− ∂+ ρ = (1 + h̄)∂− h̄(∂+ u+ − λ2 σ − )P̄ −1 − (1 + h̄)(∂+ u+ − λ2 σ − )∂− P̄ P̄ −2

 
2 1 + h̄
+λ 1− .
P̄
(4.8)

From this expression the curvature R = 8e−2ρ ∂+ ∂+ ρ is easily seen to be non-

singular at the classical singularity in all the cases a) to d) discussed at the end
of section 3 (as is obvious from the fact that the metric is non-singular there) and
furthermore in case d), it is seen that there are no curvature singularities anywhere
for either sign of κ.

14
 5. Problems in Calculating Hawking Radiation

Before we calculate Hawking radiation we would like to comment on previous

calculations of this phenomenon in 2d dilaton gravity. These comments may have
a bearing on the original calculation [2] in 4d as well.

In the CGHS calculation [1], the stress tensor anomaly is added to the classical
stress tensor trace to give

N
T−+ = e−2φ (2∂+ ∂− φ − 4∂+ φ∂− φ) − λ2 e2ρ−2φ − ∂+ ∂− ρ.
6

From the conservation equation for the stress tensor the remaining components
of the stress tensor are then determined to be

f
T±± = e−2φ (4∂± ρ∂± φ − 2∂±
2
φ) + T±± ,

with the quantum (one loop) part of the stress tensor being given by

f N 2
T±± =− (∂± ρ∂± ρ − ∂± ρ + t± (σ ± )),
6

where t± are arbitrary chiral functions to be determined by the boundary

conditions. Of course in a consistent quantization ghosts have to be included and
N → N − 24 [16,3,4] and t± must be related to the ghost stress tensor [3, 4] but
we will ignore this for the moment. The usual argument then goes as follows. To
leading order, Hawking radiation may be computed by substituting the classical
solution (corresponding to the formation of a blackhole due to an incoming matter
shock wave along σ + = σ0+ ) into the quantum piece of the stress tensor T f , and
then imposing boundary conditions. In terms of the asymptotically Minkowski

15
 1
coordinates σ̄ + = λ log(λσ + ), σ̄ − = − λ1 log(−λσ − − λa ), the classical solution is

a −
2ρ = − log(1 + eλσ ), σ + < σ0+ ,
λ
a − + +
2ρ = − log(1 + eλ(σ −σ +σ0 ) ), σ + > σ0+ .
λ

f
Substituting this in T−− and demanding that the latter vanishes for σ + < σ0+
2
1
one determines t− (σ − ) = − λ4 (1 − (1+ λa eλσ − )2
). Then observing that ∂ρ, ∂ 2 ρ → 0
+
when σ + → ∞ ( IR in the Penrose diagram) we have

f N 2 1
T−− → λ (1 − a − .
24 (1 + λ eλσ )2

This determines the Hawking radiation rate at time like future infinity to be
N 2
24 λ in agreement with earlier calculations (see for instance [18]).

This calculation however neglects back reaction. This is of course true for all
previous calculations of Hawking radiation. In the original calculations [2,18] one
quantized in a fixed background metric which means that back reaction is ignored.
But there is no sensible approximation in which back reaction can be ignored. Back
reaction is of the same order as the radiation! Within the context of this toy model
and our explicit solution of it, this problem can be resolved. But before we do it
let us elaborate on this question further.

The point is that the one loop (matter) corrected theory has an action (equation
(23) of [1] ) and associated equations of motion and constraints. Aside from the
dilaton equation, these correspond to (2.8) and (2.7) and read in this notation,

cl f
T−+ = T−+ + T−+ = 0,
(5.1)
cl f
T∓∓ = T∓∓ + T∓∓ = 0.

One has to now find a consistent solution to this set of equations (and the φ
equation of motion). Such a solution will have a classical piece plus a one loop

16
quantum correction. Now in calculating T f to order h̄ it is sufficient to substitute
the classical part of the solution into it. But to the same order one should keep
the result of substituting the O(h̄) correction to the classical solution into T cl .
One should not just keep the former as Hawking radiation and ignore the latter.
cl =
In fact the classical solution, by definition, satisfies the classical equations T−+
cl = 0, so that in order to satisfy (5.1), the leading quantum correction to
0, T±±
the classical solution when substituted into T cl must give a value which exactly
cancels the value obtained by substituting the classical solution into T f . The
CGHS calculation of course agrees with the calculations involving quantization in
a fixed background, since keeping the background fixed is tantamount to ignoring
the quantum correction to the classical solution, and is of course inconsistent with
the quantum corrected equations of motion and constraint.

A related point is that the energy-momentum conservation equation and the

⋆
equation of motion for ρ make T±± chiral fields as in conformal field theory. This
is because in any conformal gauge the stress tensor conservation law (which is
a consequence of general covariance and the matter-dilaton equations of motion)
takes the form

∂± T∓∓ + ∂∓ T+− − 2∂∓ ρT+− = 0

and the equation of motion for ρ is equivalent to the first equation of (5.1), so
that ∂± T∓∓ = 0. Since this is automatically true for t it is also true separately for
the non-ghost part of the stress tensor. Now how can we identify the ”radiation”
part of the stress tensor. As we argued earlier it does not make sense to just
subtract off the ”classical” part of the stress tensor. One can subtract the classical
value of the classical stress tensor (i.e. the value when the classical solution is
substituted into it). But by definition this is zero, so we are left with the whole
stress tensor. Also as we’ve seen, T−− is independent of σ + , and hence cannot be
zero in the region σ + < σ0+ , and non-zero for σ + > σ0+ . Indeed since T in this

⋆ Indeed the theory is, as we argued earlier, a conformal field theory.

17
section is defined to include the ghost contribution (the translation is − N6 t± →
t±± ) it is zero everywhere, for that is the equation of constraint (second equation
of (5.1)).

6. A Proposal for Calculating Hawking Radiation

How then can we identify Hawking radiation? In general relativity there is a

definition of the energy left in a system which is asymptotically flat, after radiation
has been emitted for a certain time. This is the so-called Bondi mass. This is de-
fined relative to some reference static solution and must be given in asymptotically
Minkowski coordinates. So if δTµν is the first variation of the stress tensor around
the reference solution, then for a solution (static or non-static) which asymptoti-
cally approaches the static solution at future null infinity, the Bondi mass is given
as (σ̄ ± are the asymptotically Minkowski coordinates)

+ +
ZIR ZIR
M(σ̄ ) = dσ̄ δT+ = − dσ̄ + (δT++ + δT +−).
− + 0
(6.1)

+
In the above the integral is to be evaluated at the future null infinity line IR , i.e.
at σ̄ + → ∞. Now the linearized stress tensor satisfies the linearized conservation
equation

∂∓ δT±± + ∂± δT+− = 0. (6.2)

Using this we find from (6.1),

+
ZIR
∂− M(σ̄ − ) = − dσ̄ + (∂− δT++ + ∂− δT+− )
+
ZIR (6.3)
= + dσ̄ + (∂+ δT+− + ∂+ δT−− )

= (δT+− + δT−− )I +
R

18
 This equation gives the rate of decay of the Bondi mass. We may therefore
identify the negative of the right hand side as the radiation flowing out to future
null infinity.

To proceed we need the exact solutions of our quantum corrected equations of

motion (4.3) or (4.4). Once a coordinate system is chosen, these solutions are given
in terms of two unknown chiral functions u± (σ ± ) which need to be determined from
the constraint equations and the boundary conditions. As we argued in the last
section the boundary conditions that have been used in the past, do not make sense
because of the chirality of the stress tensor, so we have to proceed in an alternative
manner. Let us first impose the constraint equations (2.7).

The stress tensor calculated from (3.13) is

r
1 |κ| 2 1X
T±± = (∂± X∂± X − ∂± Y ∂± Y ) + ∂± Y + ∂± f i ∂± f i
2 2 2
i
1 X
=e−2φ (4∂± φ∂± ρ − 2∂±
2
φ + O(κe2φ )) + 2
∂± f i ∂± f i + κ(∂± ρ∂± ρ − ∂± ρ),
2
i
(6.4)
and

r q
κ 2
2
(X+Y )
T+− =− ∂+ ∂− Y − λ e |κ| . (6.5)
2

In the coordinate system in which g± are zero we have from (4.3),

r
1X i i κ 2
T±± = ∂± f ∂± f + ∂ Y
2 2 ±
i
1X 2
= ∂± f i ∂± f i + ∂± u± .
2
i

Hence the constraint equations (2.7) become,

19
 2 1X
∂± u± + ∂± f i ∂± f i + t±± = 0. (6.6)
2
i

Now we have the problem of determining the ghost stress tensor t. This, as
well as the non-ghost stress tensors T X,Y , T f , transform like connections under
coordinate transformation because of the conformal anomaly. It is only the sum
which transforms as a tensor (since the conformal anomalies cancel between the
two). Thus under a conformal coordinate transformation σ ± → σ ′± = f ± (σ ± ),

−2
∂f ±

′f f N
T±± (σ ′ ) = [T±± (σ) + Df ± ],
∂σ ± 12
−2
∂f ±

′X,Y X,Y 26 − N
T±± = [T±± (σ) + Df ± ], (6.7)
∂σ ± 12
−2
∂f ±

−26
t′±± (σ ′ ) = [t±± (σ) + Df ± ],
∂σ ± 12

where Df is the Schwartz derivative defined by,

2
f ′′′ 3 f ′′

Df = ′ − .
f 2 f′

At this point we do not know how to proceed without making an assumption

about the boundary conditions. We assume that in a preferred coordinate system,
namely one which covers the whole space (i.e. including the region behind the
classical horizon) and is asymptotically Minkowski, the expectation value of the
f
matter stress tensor T−− vanishes. These coordinates are related to the Kruskal-
Szekeres coordinates by

1 1
σ̂ + = log(λσ + ), σ̂ − = − log(−λσ − ). (6.8)
λ λ

This condition seems to correspond to Hawking’s boundary condition, and the

20
reasoning is that there should be no f -particle energy coming in from IL− . Now
from the point of view of the exact theory the total (including ghosts) stress tensor
is zero, so that it is difficult to see what objective meaning this condition has.
Nevertheless in order to be as close as possible to the original calculation, let us
impose,

f X,Y
T̂±± = 0, T̂±± + t̂±± = 0.

The latter follows from the first equation and the constraint. However it still
leaves us the freedom of choosing the separate values of the ghost and X, Y stress
2 X,Y
λ
tensors. Let us put t̂−− = α 24 = −T̂−− i.e. we have in the σ̂ frame, an arbitrary
constant influx of ghost stress energy balanced by a constant outflow of X, Y
−
stress energy. On IR there is incoming f stress energy, which following CGHS
f 2
[1] we take to be T̂++ = aλσ0+ δ(σ̂ − σ̂0 ). Then we may take t̂++ = α 24
λ
and
X,Y 2
T̂++ = −aλσ0+ δ(ŝ − ŝ0 ) − α λ24 to be consistent with the constraints.

Then by putting σ ′ = σ̂ in (6.7), we get in the σ frame,

26 − α 1 f 26 1
t±± = − +2
, T−− =− ,
24 σ 24 σ −2

f N
T++ = −aλσ0+ δ(ŝ − ŝ0 ) −
24

Using these values in (6.6) we find,

N̄
u+ =a+ + b+ σ + − a(σ + − σ0+ )θ(σ + − σ0+ ) − log |σ + |,
24 (6.9)
N̄
−
u− =a− + b− σ − log |σ − |,
24
where

N̄ = N + α − 26 (6.10)

21
.

We now need a reference static solution. This is obtained from (4.4) and (6.9)
by putting a = a± = b± = 0 in the latter;

s
2 2 + − N̄
X0 = −Y0 = (λ σ σ + log(−σ + σ − )), f = 0.
|κ| 24

This solution is in Kruskal-Szekeres coordinates, and we need to transform this

⋆ +
into the asymptotically Minkowski coordinates σ̄ ± defined by σ + = λ1 eλσ̄ , σ − =
−
− λ1 e−λσ̄ . Under a coordinate transformation X transforms as a scalar, and (since
q
ρ(σ) → ρ(σ̄) + 2 (σ̄ − σ̄ )) Y transforms as Y (σ) → Y (σ̄) + |κ|
λ + − + −
2 λ(σ̄ − σ̄ ).
Hence we have in the new coordinate system,

s  
2 λ(σ̄ + −σ̄ − ) N̄ + − N̄ 2
X0 = − e − λ(σ̄ − σ̄ ) + log λ
|κ| 24 24
r (6.11)
|κ|
= − Y0 + λ(σ̄ + − σ̄ − ).
2

This solution corresponds to the linear dilaton solution of the classical equa-
tions. To obtain the Bondi mass of a general solution which asymptotically tends
to the above static solution, we need to linearize the stress tensors around the
latter. From (6.4) and (6.5) we have using (6.11),

s s
2 h + −
i 2 N̄
δT++ + δT+− =− ∂+ λeλ(σ̄ −σ̄ ) (δX + δY ) + λ∂+ (δX + δY )
|κ| |κ| 24
r r
|κ| |κ|
− λ∂+ δY + (∂+ (∂+ δY − ∂− δY )
2 2
.

Substituting into (6.1) we get

⋆ For the static solution these are the same as σ̂ defined above.

22
 Z
M(σ̄ ) = − dσ̄ + (δT++ + δT+− )
−

"r s
2 (σ̄ + −σ̄ − ) 2 N̄
= λe (δX + δY ) − λ(δX + δY ) (6.12)
κ |κ| 24
r r #
|κ| |κ|
+ λδY − (∂+ δY − ∂− δY ) .
2 2 +
IR

Using (3.4) and(3.2) we find that when e2φ << 1 this expression tends (not
surprisingly) to the expression given by CGHS (equation (26) of [1]) except for the
ghost terms.

Static solutions corresponding to black holes (in the classical limit) are obtained
q
by putting a = b± = 0 and a± 6= 0. Then for σ̄ >> 1, −δY = κ2 (a+ +a− ) = δX;
+

and we have from (6.12), a constant Bondi (ADM) mass

M(σ̄ − ) = λ(a+ + a− ).

The parameters a± can be of either sign and hence we may have negative
mass solutions of the theory. Of course the classical theory has such solutions
too, but there these correspond to naked singularities, whereas here these are non-
†
singular solutions (as we argued in section 4) . However one might ask whether
it is the case that we cannot generate these unphysical solutions dynamically, by
starting with positive mass solutions, in which case we might choose to ignore
them. Unfortunately this is not the case. To see this, let us compute the Bondi
mass of the analog of the dynamic CGHS solution corresponding to the formation
of a black hole by an incoming matter shock wave, and its decay by Hawking
radiation. This solution is obtained in the σ frame by putting a± = b± = 0, a 6= 0,

† This point has been emphasized by Giddings and Strominger [19].

23
in (6.6) and substituting in (4.4). Then in the region outside the classical horizon
we transform to the asymptotically Minkowski coordinates σ defined by,
1 λσ̄ + 1 + a
σ+ = e , σ − = − e−λσ̄ − 2 ,
λ λ λ
to get,
s
" + −
!!#
2 M0 + − N̄ eλσ̄ e−λσ̄ a
X =− θ(σ̄ + − σ̄0+ ) + eλ(σ̄ −σ̄ ) − log + 2)
|κ| λ 24 λ λ λ
r
κ
=−Y + λ(σ̄ + − σ̄ − ),
2

where we have put M0 = λaσ0+ the mass of the classical black hole. Comparing
with the static solution we find,

s  
2 M0 + + N̄  a +λσ̄ − 
δX = δY = − θ(σ̄ − σ̄0 ) − log 1 + e .
|κ| λ 24 λ

Substituting into (6.12) we get,

N̄ a − N̄ λ
M(σ̄ − ) = M0 − λ log(1 + eλσ̄ ) − λ
.
24 λ 24 1 + a e−λσ̄ −

In the infinite (light cone time) past σ̄ − → −∞ the Bondi mass tends to the
classical black hole mass M0 , but at future infinity σ̄ − → +∞ one gets an infinitely
negative value.

This unphysical conclusion is equivalent to the statement that the Hawking

radiation rate does not go to zero asymptotically. This rate may be calculated
either from the left hand side, or as the negative of the right hand side, of (6.3)

dM(σ̄ − ) N̄ λ2 N̄
− −
= λ − 2
→ λ2 . (6.13)
dσ̄ 24 (1 + a e−λσ̄ ) 24

Now so far α has been kept arbitrary, but perhaps the most natural choice is
α = 26, so that (see (6.10)) N̄ = N. This choice corresponds to the decoupling of

24
the ghosts from the Hawking radiation which is positive regardless of the number of
matter fields. This agrees with the two dimensional analog of the original Hawking
result [2], as well as that of [1] asymptotically but back reaction still modifies the
σ − dependence. Unfortunately although the formalism allows this value of N̄ it
is certainly not the only possibilty. Perhaps this choice has to made on physical
grounds. It is also possible that our analysis of the Bondi mass is not the complete
quantum mechanical story, and a proper treatment would resolve both this issue
as well as the question of positivity.

7. Conclusions

What progress have we made in understanding quantum black holes, and in

particular the phenomenon of Hawking radiation, from this work?

Firstly let us stress that even if we leave aside our argument for regarding
the Liouville-like theory as the complete quantum theory, it still gives us the only
consistent treatment of the semi-classical theory (i.e. to first order in κe2φ ). As
we pointed out in section three, if we just include the leading order corrections
then the field space curvature is zero, and one immediately has a soluble semi-
classical theory. All of the above calculations are then still valid except that we
cannot draw some of the conclusions that we’ve drawn from them. Thus we can no
longer explicitly demonstrate the taming of the classical singularity, and of course
there is no need to conclude that the Hawking radiation does not stop, and that
positive mass black holes radiate into negative mass solutions. Nevertheless one
has a consistent semi-classical picture of black hole radiation and back reaction. In
particular it should be emphasized again that our remarks about the inconsistencies
associated with the usual calculation of Hawking radiation which ignores back
reaction, are valid already at the semi-classical level. To belabor the point, the
calculations with the Liouville-like theory, when interpreted in terms of the ρ, φ
variables, and considered as being valid to O(κe2φ ), are the correct semi-classical
results coming from the classical CGHS theory. In particular, they show that

25
the same semi-classical physics is obtained whatever options are chosen for the
¯ is zero at the semi-
functions h, h̄ (as discussed in section three) simply because h̄
classical level. In other words we may use the exactly soluble conformal field theory
to make the calculations, provided we interpret the result as being valid only at
the semi-classical level.

Secondly we have shown that there is a class of quantum dilaton gravity the-
ories, namely those for which the field-space curvature is zero to all orders in e2φ
¯ = 0, see discussion after equation (3.1)) whose exact quantum treatment is
(h̄
⋆
possible since they can be transformed into a Liouville-like theory. These theories
allow for the first time a complete quantum mechanical treatment (including the
effects of dilaton-graviton loops) of a theory of gravity with classical black hole
solutions. Unfortunately, as we have shown, these theories may not be physical. It
¯ 6= 0 which
is an open question whether it is possible to find a soluble theory with h̄
does not have this problem, but we believe that this is unlikely. After all as we
have explicitly demonstrated, quantum mechanics does what one expects it to do,
namely it tames the classical singularities, including the naked ones! However it
thereby eliminates the usual argument (in the classical theory) for eliminating neg-
ative mass solutions on the grounds that such spaces are not globally hyperbolic.
It is possible that the problem is not so much with the soluble class of models that
we have treated, as with the original classical dilaton-gravity theory itself, which
does not have a positive definite field space metric. On the other hand it is also
possible that the fault lies with our rather heuristic treatment of the Bondi mass
in the quantum theory, and that a rigorous quantum treatment may resolve this
issue.

⋆ The objection raised by some authors on the range of integration has been answered in
section 3. In particular for the sub-class d) there can be no objection on these grounds.

26
 8. Acknowledgements

I wish to thank Geoff Harvey, Greg Moore, and Andy Strominger, for discus-
sions. This work is partially supported by Department of Energy contract No.
DE-FG02-91-ER-40672.

REFERENCES

1. C.G. Callan, S.B. Giddings, J.A. Harvey, and A. Strominger, Phys. Rev.
D45 (1992) R1005.

2. S. W. Hawking, Comm. Math. Phys. 43 (1975) 199.

3. S. P. de Alwis, Phys. Lett. B289 (1992) 278,

hepth@xxx/9205069, Phys. Lett. B (in press); and Colorado preprint, (1992)
COLO-HEP-284, hep-th@xxx/9206020.

4. A. Bilal and C. Callan, Princeton preprint (1992) PUPT-1320,

hepth@xxx/9205089.

5. T. Banks, A. Dabholkar, M.R. Douglas, and M O’Loughlin, Rutgers preprint

(1992) RU-91-54.

6. J.G. Russo, L. Susskind, L. Thorlacius, Stanford preprint (1992) SU-ITP-92-

4, and L. Susskind and L. Thorlacius, Stanford preprint (1992) SU-ITP-92-12.

7. S. P. de Alwis and J. Lykken, Phys. Lett. B269 (1991) 264.

8. F. David, Mod. Phys. Lett. A3 (1988) 1651, J.Distler and H. Kawai, Nucl.
Phys. B321 (1989) 509.

9. A. Chamseddine, Phys. Lett. B256 (1991) 379, Phys. Lett. B258 (1991)
97,and Nucl. Phys. B368 (1992) 98; J.M. Lichtzier and S.D. Odintsov, Mod.
Phys. Lett. A6 (1991) 1953; A Chamseddine and Th. Burwick, preprint
(1992) hep-th@xxx/9204002; J. Russo and A. Tseytlin, SU-ITP-2, DAMTP-
1-1992.

27
10. K. Hamada, preprint (1992) UT-Komaba 92-7, hep-th@xxx/9206071; A.
Mikovic, preprint (1992) QMW/PH/92/12, hep-th@xxx/9207006.

11. M. Green, J. Schwarz, E. Witten, Superstring Theory, Chap. 3, Cambridge

University Press (1987).

12. D. Friedan, Phys. Rev. Lett. 45 (1980) 1057. C. Callan, D. Friedan, E.

Martinec, and M. Perry, Nucl. Phys. B262 (1985) 593. E.S. Fradkin and
A. Tseytlin, Phys. Lett. B158 (1985) 316; Nucl. Phys. B261 (1986) 413. S.
Das and B. Sathiapalan, Phys. Rev. Lett. 56 (1986) 1511. R. Akhoury and
Y. Okada, Phys. Lett. B183 (1987) 65. C. Itoi and Y. Watabiki, Phys. Lett.
B198 (1987) 486.

13. A. Strominger, private communication.

14. E. WittenNucl. Phys. B311 (1988) 46.

15. T. Curtright and C. Thorn, Phys. Rev. Lett. 48 (1982) 1309; E. Braaten,
T. Curtright, and C. Thorn, Phys. Lett. B118 (1982) 115; Ann. Phys. 147
(1983) 365; E. D’Hoker and R. Jackiw, Phys. Rev. D26 (1982) 3517.

16. A. Strominger ITP Santa Barbara preprint (1992) UCSBTH-92-18,

hepth@xxx/9205028.

17. J.G. Russo, L. Susskind, L. Thorlacius, Stanford preprint SU-ITP-92-17.

18. N.D Birrell and P.C.W. Davies ”Quantum Fields in Curved Space” (Cam-
bridge 1982) and references therein.

19. S. Giddings, and A. Strominger, Santa Barbara preprint (1992) UCSBTH-

92-28, hepth@xxx/9207034.

28